Some sites have started to tell moderators to ban users using AI to generate content and reputation.

Will the community begin a machine-learning battle against the use of AI in Stack Exchange to help moderators detect cheaters (as in online chess sites)?


How can we identify ChatGPT-generated posts?

  • 7
    The ban on the use of AI-generated answers on Stack Overflow is not a fight or dispute between humans and machines. The ban is due to the technological limitation of the AI tool. Transformer models are excellent text assemblers, but they fall short in contextual evaluation and have no ability to arbitrate or judge content stored in their database, which often results in well-written answers about incorrect or even harmful content. Commented Mar 17, 2023 at 6:36
  • Possible duplicate: The future role of Stack Exchange vs. emerging AIs
    – cocomac
    Commented Mar 17, 2023 at 6:57
  • @starball I removed the complementary biased information to focus on my question.
    – user1242306
    Commented Mar 17, 2023 at 9:41
  • 1
    To answer the question as now written- whether a network site's community wants to do this (when in-line with that network site's policies on AI-generated answers) is up to the community members of that network site. And as you know, not all community members care to contribute to site curation in the same ways. There are plenty of different ways to help curate a site.
    – starball
    Commented Mar 17, 2023 at 18:20
  • I wonder why is this question closed as off-topic. We need to know this to propertly adapt to AI use.
    – user1242306
    Commented Mar 18, 2023 at 9:18

2 Answers 2


To simply answer the question in the title (and what I think your overall question is):

There are tools that exist to detect AI-generated content, although the specifics/details aren't public. They, of course, can have some limitations (although to make it more difficult to evade the ChatGPT/AI-content ban*, the specific limitations and techniques/tools used also aren't public AFAIK). There's really isn't more information that I can give you beyond that.

Of course, there are various things that users may try to attempt to do in order to evade the AI-content policies of various sites, but the moderators are quite good at finding rule-breakers.

Personally, I think that those that break the rules get dealt with and that AI will not "beat" SE.

There are a bunch of people who wish ChatGPT wasn't banned on SO (and some other sites too). However, while ChatGPT can produce answers that appears to be correct, they're often wrong in subtle ways.

Even when they do look convincing at a glance, there are still a number of ways they can be identified, so I would strongly discourage anyone from posting AI-generated answers on Stack Exchange, even more so on sites where it is explicitly banned (I'm not aware of any sites that currently explicitly allow it).


Well - part of the problem with the use of these tools is well - while they are correct linguistically, they are often factually wrong. Part of the problem here is people literally ask ChatGPT the answer, post it exactly as it is, and figure they can get away with it.

People put too much faith in ChatGPT. It doesn't reason, and it doesn't think. It’s literally a very clever language model that's somewhat better than infinite monkeys.

I'd say if someone figured out a large language model that worked perfectly for what SE does, the sharing and synthesis of knowledge, we might become obsolete.

I'd note that it's not the 'sites' that have told moderators to ban ChatGPT. Moderators banned ChatGPT because people keep posting obvious garbage on our sites, and it's a lot of unnecessary work for no benefit to anyone.

If you're not willing to specify out and ask your own questions, or are unable to answer them, Q&A isn't really for you. If you trust ChatGPT to answer your questions accurately and for certain values of truthfully, you don't need the network. If it is not even smart enough to detect its own writing, how do you expect to actually generate something sensible?

Literally most of the folks who are posting things off ChatGPT are 1) trying their luck 2) cynically hoping people won't notice the answers are a little off 3) don't realise ChatGPT writes a certain way or some combination of the above.

I don't even use a ChatGPT tool except to confirm, and often I know just from how it's written. Quite often where I'm an SME, it makes no sense.

And if you're going to fix up the output, and prompt an engineer, and independently check the answers it gives, you might as well write an answer.

All in all, there's no real place for AI generated answers here - either because they suck, or if AI worked well enough, we'd be obsolete.